|
|
Non in dialectica complacuit Deo salvum facere populum suum |
--St. Ambrose (1), ca. 400
Good evening, ladies and gentleman, members of the Association for Computational Linguistics and guests, colleagues and friends. I am honored to address you in the presidential speech on the occasion of the fortieth anniversary of the organization at this, the world's leading university in our field, and in this most attractive setting. I thank you sincerely for this opportunity.
I had first planned to talk to you about the association, how it functions and how we're making the world a better place to compute linguistically. I didn't have in mind organizational charts and association by-laws-you know the essentials of that stuff, you can read the rest of it on the ACL web site, and I've never been sure I understood all of it, anyway. No, I was aiming at a more intuitive level, more about how we do what we do, how much is compulsion and how much healthy industry, what the group dynamics are like, etc. This was developing into a John Grisham-style probe into the social and organizational fabric of our guild, how it symbiotically feeds on the ambitions but also the latent fears and desires of its members, and, naturally, how the venerable traditions of the guild serve to check and curb the most egregious excesses.
As a matter of course, in the sort of incisive expose I started, you have to include a compelling portrait of the association's executive committee and all it's machinations --the heady atmosphere surrounding the hub of technical innovation; the corridors of power at the ostentatiously expensive ACL headquarters; captains of industry, currying favors in return for well-paid consulting gigs; movers and shakers making and breaking deals with overfunded governmental agencies; the temptations of the admiring young protégées; and the late-night calls from a well-meaning Nobel committee in Stockholm, desperate to get up to speed in natural language processing. And of course, I would not neglect the overlays of power politics, personal vendetta, and sexual intrigue.
Naturally I contacted an agent to develop the film rights, and that may have been the mistake. Sydney Pollack had agreed to direct, and several prominent Thespians expressed interest --Ed Harris, to play the outgoing ACL president, Vanessa Redgrave, in the role of the ACL office administrator, Brian Brown, as the ACL VP, and Sally Field, to play the ACL Treasurer. But the rest of the executive committee applied for an court injunction, and it arrived this morning. No hard-boiled roman a cléf based on the exec! And since this arrived shortly after a meeting in which major issues were credits on the title pages of workshop proceedings, the publication status of poster sessions, and how to approach the European Union for a publication subsidy, it may be better to shun the limelights. Maybe lawyers are more interesting people for Hollywood drama, after all.
We aren't treated well in popular portrayals. I don't know of any popular portrayals of computational linguists, but if we're basically linguists -- and I believe that Computational Linguistics is basically a branch of Linguistics -- then we have some work to do on our public image. Aldous Huxley caricatured a linguist in Point Counter Point as someone who "insisted on pronouncing 'chocolate' in the original Aztec", and Nabokov's linguist colleagues in Pnin (presumably modeled after colleagues in language and literature at Cornell) spent their time trying to teach French and Russian via phonemic analysis. The major intellectual skill of Jeremy Cook, the linguist hero of David Carkeets's The Whole Catastrophe was to construct example sentences fitting unlikely descriptions, in one case a sentence with five prepositions in a row. So in popular presentations linguists are presented as pedantic, pointless and preposterous.(2)
So we'll have to leave the PR to others. This cobbler will need to stick to his last, and let you know why you--the ACL member--is working so hard. It's the same question Jake Geddes posed to Noah Cross in his orchards overlooking 1930's Los Angeles in the Roman Polanski movie, Chinatown. Jake knows that Cross is involved somehow in the death of his partner, and that lucrative business opportunities played a large role, and Jake asks Cross:
"Why do you do it, Mr. Cross, why all the bother? You already have lots of cars, houses, farms, and land. What will you buy that you can't already afford?"
John Huston, playing Noah Cross, replies to Jack Nicholson, playing Jake Geddes:
"The future, Mr. Geddes, the future."
Your own future may be very different given the proposal now under consideration by the ACL executive committee concerning program policy at future ACL meetings. The current policies effectively result in us comparing submissions from a single subfield and selecting for presentation and publications those judged by recognized experts in that subfield to contribute scientifically--through covering more data with greater accuracy, through more efficiency, or by suggesting novel or improved applications. We're concerned that our conferences, and by extension our publications, may be pandering too slavishly to too narrow a sense of scientific quality, and that some of you may have gotten the short end of the stick in all this. So in the future we want things to change. There has to be room for ideological tenacity, polemical creativity, and protection from unconstructive expert scrutiny.
The most important innovation we have in mind are "birds of a feather" sessions where groups and also refereeing are determined more by shared philosophical conviction than by the domain of study. So instead of having your statistical parsing paper evaluated by a parsing expert who may be unattuned to your empiricist ideals, in the future the evaluation will be done by a card-carrying empiricist. An inspiration for this step is the identification of many researchers with philosophical designations such as 'empiricist', 'rationalist', and 'pragmatist', but there are more possibilities. And this innovation should lead to further substantial changes in the way we select and recognize quality work.
The rationalist, or theory-driven crowd, poses a difficult problem at first, resisting as always simple quantitative summaries of the complex, multi-dimensional problems of language analysis. We first approached the problem of evaluating this group with a theorems/axioms ratio in mind, but this was rejected by the purists, who wished to claim all the infinite consequences of their axioms as their own theorems. No, the key to this group is their goal of explanatory depth, i.e., the deductive structure they impose on their material, henceforth to be expressed in explicit formal derivations, e.g., in Gentzen calculi. Since we really agree that explanatory depth is paramount, we'll operationalize the notion of quality among this group by derivation length, normalizing for the definition-length of extra-logical predicates and correcting for the deceptive triviality of the results. To keep the playing field level, we'll focus on one analytical problem at a time, e.g., pied-piping in parasitic gap constructions in Fiji. The success of this innovation depends crucially on electronic publication, since the printed page imposes an undesirable ceiling on quality (derivation length). We look forward to substantially increasing the proportion of papers devoted to long object-language proofs.
Gratian reminds us that we need to accommodate the illogical as well:(1)
Sed non in dialectica complacuit Deo salvum facere populum suum.
But it was not by means of logic that God deigned to redeem his people.
And indeed, the ACL has a tradition to uphold in this respect as well.
To continue with the empiricists, or, as they're known internally, the "just-the-facts-ma'am" group, our innovation toward philosophy-driven assessment could substantially simplify the entire process of refereeing and evaluating. Today's empiricist papers are dominated by tables in which success rates for one or another processing task are provided in 95%-confidence intervals for a combination of system parameters, preferably calculated to three decimals places beyond significance. The prose between the tables throws a sop to the older "logic jocks" customs, a custom more honored in the breach. One of the beauties of the new regime can be seen in its application to the empiricists' work. Once you've decided that all you care about is the facts, then we just have to select those papers that report the most facts. So in the future, refereeing will become much simpler-we'll focus on those papers whose tables have the greatest total cell count. A big sigma over the number of rows times number of columns, and who needs a top-heavy program committee?
A lesser group might have been satisfied with these modest changes, but the subcommittee I had the honor of chairing recognized that the simple division into empiricists and rationalists left little room for important alternative traditions in the field. To begin with themselves, the subcommittee felt that a special track was needed for unselfish facilitators of scientific progress, toiling thanklessly pro bono publico, sacrificing countless hours that would otherwise be devoted to trail-blazing discoveries of tremendous moment. So we decided that it would be best to institute a track for ACL "apparatchiki", and for whom acceptance would depend on the number of policy statements quoted, with special points for as-yet unpublished memos on research strategy from national and international funding agencies. More difficult to operationalize is the "vision thing", especially whether a paper presenter is likely to keep a straight face while conjuring up completely implausible futuristic scenarios of talking egg-beaters, dialogue-capable lawn-care tools, or toilet flush mechanisms that distinguish speech from other Bernoulli-induced sound waves. We're still working on this.
Continuing with the philosophies inspiring computational linguists, it has become clear that the association is also home to those who'd deny the presupposition in that most profound of questions in contemporary philosophy of science:
"If you're so smart, why ain't you rich?"
Our proposal to accommodate vulgar pragmatism is to suggest the creation of a "grantsmanship" track in future meetings of the association. It became clear in deliberations that the "grantsmen" constitute a separate subculture in which papers are never delivered without extensive statistics on mega-dollar budgeting, kilo person-years of effort, the names of Fortune 500 CEO's or cabinet-level officials who attend ceremonial unveilings and kick-off meetings, or what percentage of the GNP of a small German state was devoted to the project being described. Clearly this track needs a yardstick that is adjusted for inflation, and in recognition for the unparalleled achievements of our Saarbrücken colleagues, we propose that grantsmanship quality be measured in mvm's or "milli-verbmobils", which is one thousandth of the Verbmobil budget. This has the advantage of being easily convertible to person-years: one mvm is roughly a person-year.
Naturally, we could differentiate further. We have proposals for an hacker's track, "techies", where the quality formula so far incorporates inflated characteristics of system coverage and inverses of processing time and memory requirements as well as the number of CPU/OS combinations for which processing figures are available. Papers involving implementations that are possible in standard programming languages, i.e., without direct bit manipulations, are ineligible for this track.
More controversial is the proposal for an historical track, including reflections on mythic characterizations of technology, e.g., Ovid's keen eye on Daedelus's ambivalence about the techné that in its beta-version sent his son Icarus to the first recorded system crash, or yet another definitive indictment of the Turing test as mechanistic, specio-centric or unfair to the tongue-tied. Authors from electrical engineering departments would receive bonus points in evaluation here.
We haven't progressed as far with rationalization of the track for software demonstrations, but reviewing recent sessions, it's clear that we haven't cultivated bombastic self-aggrandizement as consistently as we might: we seek the scientific equivalent of King's College choir accompanied by the St. Martin in the Fields orchestra performing "Mary had a Little Lamb," and we haven't always maintained this level.
We're even including a "fantasy track" for the sorts of papers that seem to begin:
"suppose quantum effects in massively parallel computation resulted in depolarization and subsequent disorientation. Then communication would provide the necessary redundancy for computational beings."
Publication decisions will proceed in two steps: initially, all such papers will be rejected. Those authors who respond noting the treatment of misunderstood geniuses such as Galileo, Galois and Prince Myshkin will have their papers accepted in a second round.
I know that you're as enthusiastic about these possibilities as I am, and we want to enlist your active help in designing the perfect professional society of the future. Unfortunately, I'm finishing up my term as ACL president soon, so you should address your correspondence on these matters to my successor, Prof. Mark Johnson of Brown University.
I've appreciated your attention. Good night.
John Nerbonne,
ACL Presidential Speech,
University of Pennsylvania
July 9, 2002
| 1. | St. Ambrose De Fide ad Gratianum Augustum, Chap.5, Para. 42. One colleague challenged me on the identification of Ambrose as the source of this quote, sending a reference he found electronically to Gratian's Decretum (about 1140), Concordia Discordantium Canonum ac Primum de Iure Naturae et Constitutionis, VI. Item ex responsione Adriani Papae ad Carolum, c. 49. See Gratian: Text und Images der Edition Friedbergs (1879) available at the Bayerische Staatsbibliothek, and I admittedly knew the quote, not from my extensive reading in Ambrose, but only from Newman's Apologia pro sua vita. Originally charmed by this example of how even traditional scholarship might be improved by electronic methods, I checked Newman's reference, just to be sure. Result: No clear victory for electronic methods this time, I'm afraid. Newman's reference was correct! |
| 2. | And if you enjoy keeping up with artisitic misconceptions of linguistics, then Pascal Mercier's Perlmanns Schweigen is worth a read---but not because the activities of the linguist hero in mid-life crisis have anything to do with linguistics. It's a pretty good read about a scholar posing as linguist whose writer's block is nearly fatal. |
